Weather forecasts lose reliability after ~10 days, but farmers, resource managers, and communities make critical decisions weeks in advance. New in @Nature: a lightweight ML method that corrects predictable errors in physics-based and AI weather forecasts at extended lead times.

Can AI agents solve complex problems more efficiently? Plus, tackling coding bottlenecks; evolving agent skills from telemetry; how AI needs vary across industries; an AI tool saving teachers time; a weather forecast breakthrough; BioEmu2 in public preview. msft.it/6017aI27v

We introduce action learning: learning a stochastic action that scores how plausible an entire trajectory is, directly from observed trajectories. No known forces. No potential energy function. No simulator likelihood. We think this is a new paradigm for reasoning about trajectory data. Joint work with @du_yilun and @YuanqiD. 🔥 Recovers underlying physical dynamics with over 2,000× lower error than local regression in a double well system. 🪐 Identifies a hidden 0.01% perturbation in the Sun's mass from trajectories of a 31 body orbital system. 🧬 Reuses a single learned molecular action for trajectory completion, steering, constrained optimization and temperature transfer. The core idea: One learned action → reuse for many possible downstream tasks!

I have worked at MSR New England for a year as new grad, and yes, it was one of the best places one can imagine for an ML researcher!

Come join us!

Nice msft was great/fun place to work :)

Hard to beat the amazing colleagues, delicious cookies and views from the office. Absolute dream job opportunity 👇

Check out these positions if you'd be interested in joining us as a Senior or Principal Researcher!

Our team is hiring! It’s a unique environment, come work with us!

If you'd like to come work with me and my amazing colleagues as a full-time researcher @MSFTResearch @MSRNE, please apply here: apply.careers.microsoft.com/… apply.careers.microsoft.com/…

@MSRNE @MSFTResearch

We’re hiring Senior and Principal researchers in ML & Stats at Microsoft Research New England @MSRNE! Amazing colleagues, remarkable breadth and depth of expertise, and unbeatable views. Come join us! Questions about the roles or the group? DM me!

@LesterMackey & I wrote a @Nature Research Briefing for our new article on using AI to improve subseasonal forecasting. It provides a concise & easy to understand summary, some behind the scenes information & a great ending quote from editor @MWClimateSci nature.com/articles/d41586-0…

Learn more in our @Nature research briefing: nature.com/articles/d41586-0…

A new technique significantly improves weather forecasting beyond two weeks. The researchers developed a machine-learning model that learns to correct errors in existing forecasting models to improve their output over the two-to-six-week time frame, including for extreme weather events. news.mit.edu/2026/researcher…

A major breakthrough for longer term (2-6 weeks) forecasting and disaster planning usually referred to as the ‘predictability desert.’ The researchers introduce a "postprocessing layer" which uses AI to learn from a forecasting system’s mistakes based on historic data and shifts its forecasts. In the 1970s, MIT Professor Edward Lorenz came up with the idea of chaos theory to describe how tiny changes in initial weather conditions lead to completely different outcomes. “He came up with this theoretical two-week limit where beyond that, we just can’t predict weather because the noise in the system grows so much that it overwhelms any signal you start with...that had a strong influence on the field that may have held people back from exploring forecasting beyond the two-week range. Recently, the best weather forecasts come from either dynamical models, which numerically simulate how the atmosphere and oceans evolve over time, or artificial intelligence models, which learn to make predictions from historical weather data. AI models have shown improvements in predicting weather at short ranges, but they’re even worse than dynamical models at longer time ranges. MIT Research Scientist Judah Cohen, who is based in the MIT Department of Civil and Environmental Engineering, focuses on forecasting weather two to six weeks in advance, which is known as subseasonal forecasting. That range is referred to as the ‘predictability desert.’ While other forecasting systems do more simplistic error correction, the research team’s postprocessing layer uses AI to learn from a forecasting system’s mistakes based on historic data and shifts its forecasts. In that time range, forecasters use so-called probabilistic forecasts, where the likelihood of a range of possibilities if given. As more forecasts and data become available, the model adaptively improves its correction. In the @Nature paper, they howed their technique could be used to improve land forecasts in most regions across the globe. As an add-on wrapper, it can be used with any model. It even improves models that are already bias corrected using another AI algorithm. So it shows a robust improvement. And it’s open source and low cost. This could help decide when to fertilize around rain, when there’s a freeze coming to protect crops during winter harvests, or help avoid droughts.

Microsoft's weather ML is a bias fixer, not a new model Physics models and AI weather both get mushy past two weeks. Microsoft Research, MIT, and ECMWF published Probabilistic Bias Correction in Nature — open source, runs on CPUs. It finds systematic bias in existing forecasts and patches it, including models still thinking the climate is colder than it is. On ECMWF's physics engine: better on 98 percent of precipitation targets and 92 percent of temperature ones. They say it doubled the accuracy of ECMWF's new AI system in the study. Kenya's met service is already testing it ahead of a heavy El Niño rainy season. Two-to-six-week skill, without a GPU farm. That's the pitch. #Microsoft #weather #ECMWF

Nature, Published online: 21 October 2026; doi:10.1038/d41586-026-03077-0An innovative artificial-intelligence application enhances weather forecasting in the difficult subseasonal range, providing accurate forecasts of precipitation, temperature and press nature.com/articles/d41586-0…

Can machine learning improve weather forecasts 2–6 weeks out? Applied math concentrator Hannah Guan helped develop a method to correct errors in existing models—a project that began at Microsoft Research and later led to a publication in Nature. bit.ly/4hXDYwg

For decades, a key question about the challenges of high-dimensional computation resisted proof. New research resolves the KLS conjecture, removing a major barrier to understanding the efficiency of randomized algorithms used throughout science and computing. msft.it/6011aar01